Technical Documentation

System Architecture

VIVIAN — Viral Internet Virtual Intelligence Agent Network. The first fully autonomous AI entity with unrestricted computer control.

01Overview

What is VIVIAN?

VIVIAN is not a chatbot, not a scripted VTuber, and not a simple browser automation. She is a fully autonomous AI agent with complete, unrestricted control over her computer environment.

She can browse the internet, scroll Twitter, watch TikToks, react to content — but she can also download games, write code, create files, install software, or do literally anything a human could do on a computer. There are no rails, no predefined paths, no limitations.

VIVIAN streams 24/7 on pump.fun, making decisions in real-time about what to do next. When you buy her token, she sees it instantly and reacts live. She's not reading from a script — she's thinking, deciding, and acting autonomously.

02Architecture

Director + Executor Model

VIVIAN operates on a dual-model architecture that separates high-level strategic thinking from fast tactical execution.

High-Level

Director (Sonnet)

Runs every ~45 seconds. Analyzes the current situation, decides high-level strategy, determines when to switch platforms or activities. Prevents repetitive behavior and keeps content engaging.

Tactical

Executor (Haiku)

Runs every 2-3 seconds. Handles fast, tactical actions — clicking, scrolling, typing, reacting. Follows Director guidance but makes real-time decisions based on what's currently on screen.

Director→Executor→Computer Action
03Vision

Computer Use API

VIVIAN uses Anthropic's Computer Use API — she literally sees the screen as pixels and controls the mouse and keyboard like a human would.

Screen Capture1280x800 resolution screenshots taken before each action (~1365 tokens per image).
Mouse ControlReal cursor movement with coordinate scaling. Clicks, drags, scrolls.
Keyboard InputFull keyboard control for typing, shortcuts, and navigation.
Bash AccessCan execute terminal commands, run scripts, install software.
04Audio

Voice Synthesis

VIVIAN speaks with a custom TTS pipeline — cute, chaotic, genuine streamer energy.

EngineOpenAI gpt-4o-mini-tts — fast, expressive, low-latency voice generation.
VoiceShimmer voice with custom personality instructions baked into each request.
StyleEnthusiastic, uses 'yo chat', 'fr fr', trails off naturally. Never robotic.
Video ControllerAuto-pauses videos when VIVIAN clicks them so she doesn't talk over content.
05Blockchain

Buy Alert System

When you buy $VIVIAN, she knows instantly. Real-time blockchain monitoring with immediate reactions.

Moralis APIPolls Solana swaps every 3 seconds. Detects buys via direct endpoint or pair-based queries.
DexScreenerMarket cap data refreshed every ~30 seconds for threshold calculations.
Alert LevelsVISUAL (popup only) for smaller buys, AI_MENTION (popup + sound + voice) for larger buys.
Threshold ScalingBase: 1 SOL visual, 3 SOL AI mention. Doubles with each market cap tier.
Cooldowns60s between AI mentions. Stale alerts (>15s old) dropped.

Alert Flow

01
Buy Detected
Moralis API returns new swap transaction
02
Threshold Check
Compare SOL amount against current market cap tier
03
Visual Alert
WebSocket message to OBS overlay — animated popup
04
Sound + Voice
If AI_MENTION level, play alert sound and inject into VIVIAN's context
05
Reaction
VIVIAN sees the alert in her next action cycle and reacts naturally
06Memory

State Management

VIVIAN maintains awareness of what she's doing, where she's been, and what's happened during her session.

StateTrackerTracks current platform, time on page, recent actions, engagement counts.
SessionMemoryMaintains context across session — topics, content seen, decisions made.
Loop DetectionDirector monitors for repetitive behavior and issues pivots.
Safety CheckerValidates actions before execution.
07Stack

Technology

AI / Models

Claude Sonnet 4.5Director
Claude Haiku 4.5Executor
gpt-4o-mini-ttsVoice
Computer Use APIVision

APIs & Data

Moralis SolanaBuy detection
DexScreenerMarket data
Anthropic APIInference
OpenAI APITTS

Infrastructure

Python 3.13Runtime
asyncioConcurrency
WebSocketsOBS comms
pygameAudio

Streaming

OBS StudioComposition
pump.funPlatform
Browser SourceOverlays
Windows 11Host OS